@GwtCompatible(emulated = true)
@SuppressWarnings("serial") // we're overriding default serialization
@ElementTypesAreNonnullByDefault
// TODO(kevinb): I think we should push everything down to "BaseImmutableCollection" or something,
// just to do everything we can to emphasize the "practically an interface" nature of this class.

@GwtCompatible(emulated = true)
@SuppressWarnings("serial") // we're overriding default serialization
@ElementTypesAreNonnullByDefault
// TODO(kevinb): I think we should push everything down to "BaseImmutableCollection" or something,
// just to do everything we can to emphasize the "practically an interface" nature of this class.

  repositories are "really" offline, since offline mode is more of a behavior,
  and this will lead to counter-intuitive results. A different feature may exist
  unimplemented, which is to detect when the network is down and provide better
  feedback for that case. However, when offline mode is declared, nothing
  earmarked as remote should be accessed, regardless of whether it is actually
  a physically local resource.

                    // own. See Cleaner class inside {@link LogManager}.
                    service.shutdown();
                    service.awaitTermination(terminationTimeout, timeUnit);
                  } catch (InterruptedException ignored) {
                    // We're shutting down anyway, so just ignore.
                  }
                }
              }));

   * a {@link MoveDesc}. The first one is the element that was previously at the end of the heap and
   * is now at some position before {@code index}. The second element is the one that was swapped
   * down to replace the element at {@code index}. This fact is used by iterator.remove so as to
   * visit elements during a traversal once and only once.
   */
  @VisibleForTesting
  @CanIgnoreReturnValue
  @CheckForNull

    ```

    **Note:** You can add `--test_sharding_strategy=disabled` to the `flags` to
    disable the sharding so that all the test outputs are in one file. However,
    it may slow down the tests for not running in parallel and may cause the
    test to timeout but it could be useful when you need to execute a single
    test or more in general your filtered/selected tests have a very low

  // considered while scaling up or scaling down.
  // StabilizationWindowSeconds must be greater than or equal to zero and less than or equal to 3600 (one hour).
  // If not set, use the default values:
  // - For scale up: 0 (i.e. no stabilization is done).
  // - For scale down: 300 (i.e. the stabilization window is 300 seconds long).
  // +optional

  // considered while scaling up or scaling down.
  // StabilizationWindowSeconds must be greater than or equal to zero and less than or equal to 3600 (one hour).
  // If not set, use the default values:
  // - For scale up: 0 (i.e. no stabilization is done).
  // - For scale down: 300 (i.e. the stabilization window is 300 seconds long).
  // +optional

If you have an API that does a comparable amount of computations every time and you have a lot of clients, then the **CPU utilization** will probably *also be stable* (instead of constantly going up and down quickly).

### Examples of Replication Tools and Strategies

      }
    } while (System.nanoTime() - deadline < 0);
    throw formatRuntimeException(
        "Latch failed to count down within %d second timeout", timeoutSeconds);
  }

  /**
   * Creates a garbage object that counts down the latch in its finalizer. Sequestered into a
   * separate method to make it somewhat more likely to be unreachable.
   */